This repository encompasses a pipeline for building boosted regression tree (BRT) patch distribution models using NOAA EcoMon abundance data and Copernicus Marine Data Store oceanographic covariates
Output predictions are daily, 1/12° resolution predictions of a species patch from 1993 to 2019 within the Northeast Atlantic.
In addition, the real_time_forecasts folder hosts a
Shiny application template for delivering real-time daily product.
Example output: Probability of a dense C. finmarchicus patch daily from April 1st to June 1st, 2015.
New model versions can be customized in the yaml_maker.R
file. Configuration allows for choice of a species; patch threshold;
source column within the ecomon::scale_ecomon()
output or alternative custom dataset; set of selected copernicus and
derived variables; and hyperparameter customization.
Steps:
data_preparation/generate_base_datasets.R accordingly.yaml_maker.R and execute.create_version.R and
execute.mnt/ecocast/projectdata/students/ojohnson/copernicus/predict_version.R and
execute within a shell. Multiple species/version pairs can be run
through